Task Manager MCP Server

Task Manager MCP Server

This MCP server allows agents to manage tasks, including creating tasks with uncertainty areas, updating task dependencies and uncertainty resolution, and transitioning task status through their lifecycle. It provides structured task management capabilities for agents working on complex multi-step problems:

• Create and organize tasks in hierarchical structures with dependencies
• Update tasks with dependency information and uncertainty area tracking
• Track task progression through defined states (not-started, in-progress, complete)
• Manage uncertainty areas by identifying and tracking areas that need resolution before task execution
• Orchestrate workflows with proper dependency validation

The tools have been tested extensively and successfully with VS Code and Claude Sonnet 4. (GPT-4.1 and GPT-5 do not seem to work very well unfortunately.)

Note: When using this MCP server, you should disable the Todo List tool in VS Code.

Tools

1. create_task
   • A tool to create a new task
   • Inputs:
     • title (string): A concise title for this task
     • description (string): A detailed description of this task
     • goal (string): The overall goal of this task
     • dependsOnTaskIDs (array of strings): Task IDs this task depends on; referenced tasks must already exist
     • definitionsOfDone (array of strings): A detailed list of criteria that must be met for this task to be considered complete
     • uncertaintyAreas (array of objects): Areas requiring clarification, each containing:
       • title (string): A concise title for this uncertainty area
       • description (string): A description of this uncertainty area
   • Behavior:
     • Creates one separate task per uncertainty area (status not-started), and links those tasks as dependencies of the newly created main task
     • Uncertainty area tasks are created as readonly tasks with titles like "Resolve uncertainty: {area.title} for task: {taskTitle}"
     • Any provided dependsOnTaskIDs are also applied as dependencies of each uncertainty-area task
     • Only one task can be in-progress at a time across the entire task tree
   • Returns: Confirmation including a list of tasksCreated (uncertainty-area tasks first, then the main task) and executionConstraints with constraint information when applicable
2. update_task
   • A tool to update an existing task
   • Inputs:
     • updates (array of objects): A list of updates to apply to existing tasks, each containing:
       • taskID (string): The identifier of this task
       • newDependsOnTaskIDs (array of strings): A list of additional task identifiers this task depends on
       • newDefinitionsOfDone (array of strings, optional): A detailed list of additional criteria that must be met for this task to be considered complete
       • newUncertaintyAreas (array of objects): A detailed list of additional areas where there is uncertainty about this task's requirements or execution, each containing:
         • title (string): A concise title for this uncertainty area
         • description (string): A description of this uncertainty area
   • Behavior:
     • If newDefinitionsOfDone are provided, they are added to the task's existing definitions of done
     • Creates a separate task for each newUncertaintyArea (status not-started), applies newDependsOnTaskIDs to those tasks, and adds these newly created tasks as dependencies of the target taskID
     • If the target task is already complete, the operation fails
     • This operation does not update existing uncertainty-area tasks or their statuses
   • Returns: Confirmation including tasksUpdated (array of updated tasks) and, if any were created, tasksCreated, plus executionConstraints with constraint information when applicable
3. transition_task_status
   • A tool to transition the status of a task
   • Inputs:
     • taskID (string): The identifier of this task
     • newStatus (enum): The new status ("not-started", "in-progress", or "complete")
     • outcomeDetails (array of strings, optional): Required when transitioning to "complete"
     • verificationEvidence (array of strings, optional): Required when transitioning to "complete"
   • Behavior:
     • Allowed transitions:
       • not-started → in-progress
       • in-progress → complete (requires outcomeDetails)
       • complete → in-progress
     • To start (in-progress), all dependencies of the task must be complete
     • Tasks must have their uncertainty areas updated before they can be started
     • Only one task can be in-progress at a time across the entire task tree
     • Before starting a task, it must have its uncertaintyAreasUpdated flag set (done via update_task)
   • Returns: Status transition confirmation including taskUpdated and executionConstraints with constraint information when applicable
4. task_info
   • A tool to retrieve full information about a task
   • Inputs:
     • taskID (string, optional in single agent mode): The identifier of this task
   • Behavior:
     • In normal mode: taskID is required and the tool returns information for the specified task
     • In single agent mode (when SINGLE_AGENT=true): If no taskID is provided, returns information for the current in-progress task
   • Returns: The complete stored task object and executionConstraints with constraint information when applicable

Single Agent Mode

The task manager supports a special "single agent mode" that can be enabled by setting the environment variable SINGLE_AGENT=true. In this mode, the server tracks which task is currently in progress, enabling enhanced workflows for single-agent scenarios.

This is useful for long-running agents where the agent loop is compacting/summarizing the agent's conversation history to prevent exceeding the context window limit. In these cases, the agent may "forget" which tasks exist because the task IDs have been removed from the context window. Single agent mode allows the agent to use the task_info tool without needing to specify a taskID, enabling it to regain information about the current task tree.

Do not enable single agent mode if you plan to use this MCP server with multiple agents at the same time. If enabled, only one agent should use the MCP server at any one time.

Enabling Single Agent Mode

Set the environment variable before starting the server:

Or in Claude Desktop configuration:

Recommended Agent Prompt Snippets

Instructions section

Task Management tools section

Usage with Claude Desktop (uses stdio Transport)

Add to your claude_desktop_config.json:

Usage with VS Code

For quick installation, use of of the one-click install buttons below.

For manual installation, add the following JSON block to your User Settings (JSON) file in VS Code. You can do this by pressing Ctrl + Shift + P and typing Preferences: Open User Settings (JSON).

Optionally, you can add it to a file called .vscode/mcp.json in your workspace. This will allow you to share the configuration with others.

Note that the mcp key is not needed in the .vscode/mcp.json file.

NPX

Running from source with HTTP+SSE Transport (deprecated as of 2025-03-26)

Run from source with Streamable HTTP Transport

Running as an installed package

Install

Run the default (stdio) server

Or specify stdio explicitly

Run the SSE server

Run the streamable HTTP server

License

This package is licensed under the MIT license.